@inproceedings{Henriksen:2022:10.1145/3477314,
author = {Henriksen, P and Leofante, F and Lomuscio, A},
doi = {10.1145/3477314},
pages = {1031--1038},
title = {Repairing misclassifications in neural networks using limited data},
url = {http://dx.doi.org/10.1145/3477314},
year = {2022}
}

Download

TY  - CPAPER
AB  - We present a novel and computationally efficient method for repairing a feed-forward neural network with respect to a finite set of inputs that are misclassified. The method assumes no access to the training set. We present a formal characterisation for repairing the neural network and study its resulting properties in terms of soundness and minimality. We introduce a gradient-based algorithm that performs localised modifications to the network's weights such that misclassifications are repaired while marginally affecting network accuracy on correctly classified inputs. We introduce an implementation, I-REPAIR, and show it is able to repair neural networks while reducing accuracy drops by up to 90% when compared to other state-of-the-art approaches for repair.
AU  - Henriksen,P
AU  - Leofante,F
AU  - Lomuscio,A
DO  - 10.1145/3477314
EP  - 1038
PY  - 2022///
SP  - 1031
TI  - Repairing misclassifications in neural networks using limited data
UR  - http://dx.doi.org/10.1145/3477314
UR  - http://hdl.handle.net/10044/1/100460
ER  - 

Download